Few issues in the film industry today are as widely analyzed and debated as the potential and impact of AI technology. When it comes to documentary filmmaking, the debate takes on a new level of importance, as the format is often tied to journalistic notions of truth and reality. Top documentarians gathered at this year’s Amsterdam International Documentary Film Festival (IDFA), the world’s largest documentary festival, to discuss best practices, pressing warnings, and what the future holds for documentary as the use of AI becomes increasingly prevalent.
Oscar-nominated American filmmaker and investigative reporter David France, at IDFA with his Sundance-hit documentary Free Leonard Pelletier, recalled the first time he used artificial intelligence in 2020’s Welcome to Chechnya. The film chronicled the persecution of members of the LGBTQIA community in Russia’s semi-independent autonomous republic, and dealt with a very sensitive situation. In order to speak to the persecuted, France needed to ensure that their identities were not revealed. “It was a story that needed to be told, but it was difficult to tell because those who were able to escape were being pursued around the world.”
The solution France and his team came up with was to hide the characters’ faces by digitally overlaying them with other people. “Nothing about their micro-reactions or emotions changed. You could see the original human crying or laughing while using someone else’s face. At the time (2019), we didn’t call it artificial intelligence. We called it machine learning. It just seemed amazing.”
The director recruited 23 gay activists from New York to lend their faces and voices to the project in an innovative process, awarding the team behind the innovation a technical Oscar. Still, the filmmaker came under scrutiny for his use of AI. “While we were doing this, everyone was calling it a deepfake. We kept saying, it’s not a deepfake. Deepfakes are a crime and AI is a tool.”
France will once again use AI in its films, including the latest film that tells the story of activist Pelletier, who was imprisoned for half a century after being convicted. In this case, the director used AI to modify and rejuvenate Pelletier’s voice. This resource was necessary because the recordings heard in the film were obtained illegally, as Pelletier was unable to speak to journalists from prison. “Plus, Leonard went from 30 to 80, and his voice showed age.”
British filmmaker Mark Isaacs is attending IDFA with “Synthetic Honesty,” and has signed a deal with the Nominal Lab to support research into the possibility of teaching AI characters authenticity. A blend of fact and fiction, the doc was created in collaboration with Romanian actress Ilinca Manolace (Don’t Expect Too Much from the End of the World), and Isaacs expertly manipulates images through filters and other techniques to emulate what an AI-generated sequence would look like.
Isaacs conducted the experiment at Synthesia, a synthetic media generation company that develops software used to create AI-generated video content. “You can choose a character and type in what they say,” he explained. It was interesting at first, but the director said he was “bored to death” because the AI characters had “a very limited range of emotions.” “It was interesting at first, but I quickly got bored with it.” When she met Manorash at a festival in Bucharest, he suggested they work together on a film and together create the half-real, half-digital character she would play on screen. “She was more interesting and there was more variety in what we could do with her. Most actors are afraid of having their persona changed to an AI, but she really liked it.”
The director pointed out that his work was “not journalistic” and did not want to clarify what was generated by AI and what was not included in the film. “The point of this film is to raise questions about images and what’s going on with representation and the death of the camera. I didn’t want to spoil it by labeling things.”
A large portion of the conversation was devoted to understanding the impact of AI on archival footage. “For archives, the implications are very serious,” said Portuguese filmmaker Susana de Sousa Díaz (“Fordlandia Panacea”), this year’s guest of honor at IDFA and a documentarian who works primarily with archival images. “The archival status of the footage makes it much easier to contextualize. There is a risk here that not only will audiences believe fake archival footage, but also that people will stop believing in anything. In both cases, our regime of truth is completely shaken.”
“Since the transition to digital media, there has been an increased discourse about the imperfections of reality in black-and-white and low-resolution images,” she added. The director also said that working with archives is not just about what can be seen and saved through research, but also about thinking about all the gaps in materials and memory. “The question I’m interested in is actually very simple and at the same time very complex: What happens when a technology that tries to fix everything enters a field where absence itself has meaning?”
Reflecting on this contemporary challenge, Emmy Award-winning filmmaker and graphic designer Eugen Breunig (Trafficked) worked with the Archive Producers Alliance to establish a set of guidelines for best practices when using generative AI in archive-driven filmmaking. “In documentary, there is no organizing body telling you how to do things,” he stressed. “There are no laws or rules. All we can do is self-regulate and hold ourselves to certain standards in order to be responsible as storytellers, newsmakers and image-makers.”
Breunig pointed out that the most basic, yet most helpful, is for producers to create a cue sheet that lists the technology used and when and how it was used throughout the film’s production. “At some point, people are going to have questions,” he warned. It’s best to stay ahead of the curve.
At one point in the conversation, the designer played a synthetic news clip from the 1990s produced by SORA to illustrate how faithful AI-generated sequences are now. “Of course, it was possible to make fake videos before, but it required a lot of production costs and a lot of time. Now it’s too cheap and too fast,” he added.
“Trust in the media is at an all-time low,” the filmmaker warned. “That means that trust in archives is also under threat. If people start to distrust the news, which they already do, distrust of documentary filmmaking could potentially develop as well, and in some cases, it’s warranted.”
